Study On Fuel Classification And Estimation Of Fuel Load By QuickBird Image And TM Image In Tahe Area

In this paper, remote sensing images on Yangjiang River Tahe Forestry Bureau in Heilongjiang Province were applicated to research forest fuel types classification and fuel loads estimation, on the basis of summing up forest fuel at home and abroad to choose Tahe Forestry Bureau which fuel load accumulated more and vulnerable to fire for study area, Quickbird and TM images were used to perform supervised classification, study area was divided into broad-leaved forest, Pinus sylvestris forest, larch forest, Shrub, meadow, marsh, burning slash, water, roads, houses 10 categories. The results showed that the overall classification accuracy of Quickbird image was 79.35%, For TM image,the overall classification accuracy was 72.90%. Compared Quickbird image with TM image classification accuracy increased 6.45%. Mapping accuracy and user accuracy of Broad-leaved forest, Pinus sylvestris forest and larch forest were higher, these could meet the needs of this study, the case against woodland, Quickbird image's classification accuracy was higher than TM image's, the user accuracy of larch and the mapping accuracy of Pinus sylvestris were particularly evident, TM image could not be completely distinguish meadow as Quickbird image did, classification accuracies of the two images for non-fuel category were generally high,but the difference was little.choosing different image features of Quickbird and TM images: band gray values and combinations, vegetation indexes for selecting variables.multiple linear regression analysis method was chosen to establish various types of fuel load estimation models directly. Average diameter at breast height and average tree height as the intermediate variables to establish regression equations with images features,then these stand factors as independent variables to establish regression equations with fuel loads, Thus image features and fuel load estimation models were established indirectly. Utilising Quickbird and TM images,selecting average diameter at breast height and average tree height,using direct and indirect method to establish fuel load estimation models:litter fuel load, surface fuel load, air fuel load, 1h timelag fuel load, 10h timelag fuel load, 100h timelag fuel load, 1000h timelag fuel load and total fuel load.after adjusting decision coefficient(R~2),mean absolute error(MAE) , mean relative error (MRE),relative mean square error (RMSE) as evaluation indexes to evaluate all accuracies of models ,selected the best model by comparing model accuracies which established through different images, different intermediate factors,different methods. The results showed that the best method was choosing shadow fraction of Quickbird panchromatic image as independent variable, the average diameter at breast height as intermediate factor to establish various fuel load estimation models indirectly. So high-resolution remote sensing image— Quickbird image could improve the accuracies of fuel load estimation models to certain extent.